The rise of SD-WANs has raised an interesting debate. Is the internet good enough to replace a private network for an enterprise WAN?
A decade ago, no one would have even considered this, but broadband speeds have increased and more things have moved to the cloud. Also, SD-WAN technology allows for dynamic path selection, which protects the WAN from outages so companies can use multiple broadband connections instead of something like MPLS.
Global SD-WAN vendor Aryaka recently examined this question in its “State of SD-WAN Connectivity” report (registration required), which measured and compared data transport from the same pairs of locations using both the internet and over Aryaka’s own global private network. The test run was a randomly created 100 KB file, and connect time and transfer time were captured. The application response time was then calculated as the sum of these two metrics.The thesis the tests were conducted under was that for short distances, such as Boston to Washington, D.C., or London to Munich, the difference in application response times for a private network and internet would be negligible, but over long distances there would be a difference, so mostly long-haul connections were used. This would be very common to find in global multi-national companies that have offices scattered all over the globe. The tests were performed over the period of a full year and the data averaged.
Response time: Internet vs. private networks
The first interesting finding, although not surprising is that the internet has a high degree of variability in the response time. For example, the response time between Dubai and Dallas ranges from 750 milliseconds to 2 seconds (153 percent variability) where the private network variation is only 12.55 percent. Application response time fluctuation can be highly frustrating to users. With a consistent experience, users know what to expect. When it fluctuates, they have no idea if things are working, so they restart apps or resend emails and that can exacerbate the problem.
Below are the some of the measurements taken and the variability in them on public and private networks.
- Dubai to Dallas: internet – 153%, private network – 12%
- Dubai to London: internet – 181%, private network – 12%
- Dubai to Johannesburg: internet – 96%, private network – 0.25%
- Dubai to Mumbai: internet 212%, private network – 15%
- San Jose to Shanghai: internet 152%, private network – 25%
The next metric Aryaka looked at was the average response time.
- Dubai to Dallas: internet – 1.185 seconds, private network – 0.375 seconds
- San Jose to Chicago: internet - 0.194 seconds, private network – 0.158 seconds
- Dubai to London: internet – 4.24 seconds, private network – 0.19 seconds
- Frankfurt to Shanghai: internet – 1.99 seconds, private network – 0.2 seconds
- Dubai to Beijing: internet – 1.99 seconds, private network – 0.2 seconds
- San Jose to Shanghai: internet 3.97 seconds, private network 0.306 seconds
The data shows that in general, the longer the distance, the bigger the difference between application response times on a private network and internet. This makes sense given the number of hops and different networks the traffic must traverse. This is analogous to driving on a state highway versus an Interstate. If I’m going 5 miles, there won’t be much difference in time. If I’m going cross country, the dedicated freeway will get me there much faster than having to use state roads that go through towns and have traffic lights. Also, the longer the distance, the greater the variability, which has a direct impact on user experience and productivity.
The primary takeaway from this study is that despite advancements in internet speeds, the internet is still not on par with private networks on a global scale. For small companies that only need to connect regionally, the internet is fine. However, global multinational companies that have mission-critical applications need to be careful of where they use internet connectivity. It may be useful as a backup connection, but it’s not ready to replace that private connection.